L3, L4: Sensing, Actuation & Control + Autonomous Systems

what makes up a robot

- body

- sensor(s)

- effector(s)

- control mechanism

- energy/power source

mechanomorphic

machine-like

zoomorphic

animal-like, bio-inspired, biometric

anthropomorphic

human-like

morphology

body layout, degrees of freedom

degrees of freedom

equal to the number of independent parameters that define the configuration

how many degrees of freedom does a rigid object in 3D space have

6 DoF

how many degrees of freedom does the human arm have

7 DoF

why is redundancy in regard to degrees of freedom useful

facilitates optimisations such as:

- energy minimisation

- obstacle avoidance

examples of low level percepts

light level, colour, sound, temperature, texture, smell, tilt, position

examples of high level percepts

objects, people, scenes

examples of abstract percepts

intentions, meaning, affective states

external perception (exteroception)

sensory systems to monitor outside environment

internal proprioception

sensors within the body

direct perception

immediate apprehension of the environment through sensory data alone (no interpretation)

inference

using incomplete sensory inputs which is then interpreted some way

advantage of a robot sensing its own actions

it may be able to determine if its doing what its supposed to be doing

disadvantage of a robot sensing its own actions

interference

what is the control mechanism concerned with

- reasoning

- planning

- learning

- perceiving

- doing

cybernetics

the study of control and communication in the animal and the machine

principles of cybernetics

- parsimony: simpler is better (reflexes)

- exploration: never stay still except when recharging

- attraction (+ve taxis/trophism): motivated to move toward something

- adversion (-ve taxis/trophism): moves away from negative stimuli

- discernment: ability to distinguish between productive and unproductive behaviour

Hans Moravec's 'Stanford Cart'

- TV cameras took pictures of scenes

- robot planned path between obstacles

- it moved in 1 metre spurts with 10-15 min, stops for image processing and planning

- successfully crossed a room full of obstacles in 5 hours damn

Sense-Plan-Act ('SPA')

- robot senses the world and constructs a global world map

- robot plans all the directives needed to reach the goal

- robot carries out the first directive

- repeat (sensing the consequences of its action and re-planning the directives)

problems with the 'SPA' approach

- closed world assumption (hard to include everything in the world model, huge world models, hard to keep track of all changes)

- slowness (inability to act quicly)

- impractical

behaviour based robotics (BBR)

- layered reactive approach

- pioneered by Rodney Brooks MIT

- new emphasis on (simple) living examples of intelligence

self sustaining systems (homeostatic)

- make active contributions to their own persistence

- do not contribute to the maintenance of the conditions for persistence

recursive self-sustaining systems (autopoietic)

- contribute actively to the conditions for persistence

- deploy different processes of self-maintenance depending on environmental conditions

autonomous

capable of acting without constant human interaction

distributed autonomy issues

- damage/overheatinng

- communication between computer and robot could be limiting factor

- processing speed

behaviour based robotics requirements

- multiple goals (possibly conflicting)

- multiple sensors (often inconsistent)

- robustness to component failure

- extensible

subsumption architecture

multiple layers of simple behaviours (lower layers (e.g. avoiding obstacle) can override higher layers (e.g. explore))

why has subsumption been criticised

for not scaling well

what are majority of real world robots

subsumption-based vacuum cleaners

potential field

assigns a scalar value to every point in the robots environment, low points for it to move toward and high points to detract it.

cons of potential fields

- vulnerable to local minima

- susceptible to cyclic/oscillatory behaviour

- may require frequent resampling of the world

pros of potential fields

- smooth trajectories based on gradient

- no path planning (only needs the local vector)

- works well in dynamic environemnts

morphological computing

robots physical body does the computing, blurs the line between controller and to-be-controlled